Newcomer quick-start flow
- Vladimir Serjanin
- May 1
- 6 min read
Pecan is an AI predictive analytics software, designed for impact. Get accurate, actionable predictions in days and unlock the power of AutoML.
Summary
The problem we face is that newly registered users tend to disengage from additional sessions on the platform. Our goal is to find ways to keep users intrigued and help them extract value from their Pecan experience.
My input
As the Senior Product Designer, I collaborating closely with a team comprising multiple Data Analysts, two Product Managers, the Marketing Team, and a Marketing Designer, we focused on improving the user journey, particularly during the evaluation phase. Our goal was to keep users intrigued and ensure they extract maximum value from their Pecan experience. By conducting research and customizing content on landing pages, we aimed to captivate users from their initial interaction with the platform. This involved close collaboration with stakeholders to align our efforts with overarching business objectives.
How to train a model?
Connect your database: Integrate your raw database to Pecan’s secure, hosted service.
Train an ML model: Using Pecan’s SQL Editor, write queries with prediction goals and map your data.
Get live predictions: Explore the results dashboards and set up recurring live predictions.
Research
In our research phase, we collected in-depth insights about potential users and their interactions with our platform. Our aim was to pinpoint pain points in users' initial sessions and understand their origins. This exploration revealed challenges related to specific platform use cases aligned with users' business goals. To address this, we plan to shift toward a user-centric approach, focusing on clarifying loan-related issues and the necessary steps for resolution.
First moves in the platform
Most users to skip key steps in the onboarding: Many users skip the onboarding steps, leading to missed setup instructions and confusion later on.
Exploration of the home page and Demo Model: Users explore the home page, click on the pre trained Demo Model, and examine its result dashboards.
Difficulty in training a custom ML model: Users often face issues when training models, from data mapping to writing queries, leading to drop-offs before getting to results.
Users don’t interact with the provided tutorials: Even though users have access to a variety of videos and step-by-step guides, they seldom engage with them during their product evaluation stage.
What caught their eye?
To better understand our users' journey and assist them during the evaluation phase, I worked closely with the Marketing team. We researched which content captured users' attention on our landing pages and what value they seek from the platform.
Assumptions
Complex and labor-intensive field: Predictive analytics and machine learning involve intricate tasks like connecting databases and operating them using SQL code and various structures.
Confusion around what they need to do: Users often appear confused about how to get started with Pecan. They're uncertain whether they should connect data first, upload files, or get in to the SQL editor to train their predictive model.
Users want fast and easy results: New users in the evaluation stage seek quick, effortless results to determine if Pecan is suitable for their business and professional needs and if it's worth their time.
Model training dissatisfaction: Users enter the platform and attempt to train an ML model. When they encounter failure, they leave without understanding the platform’s value for their business needs.
User interviews
Demo Model is useless without knowing its data: "I don't find the dashboards helpful because I don't know what's in the Demo Data. Knowing this information is important for me to understand and use the dashboards effectively.”
Not all users can connect data or have it at hand: "I can't connect my data easily because I need special permissions. If I’m cautious, and want to use part of my data, I need to locate it and convert into a CSV file for the platform."
Comprehension around machine learning is low: "I don't fully grasp machine learning concepts, such as why I need write SQL quarries, what specific data is required, or how it works. I'm also unsure about the need for a training set in machine learning."
Summary
Lack of value of Demo or Custom Models: If users find the Demo Model unhelpful and are unable to train a custom ML model, they don’t experience any meaningful value from Pecan.
Database permissions limit model training: If most users lack permissions to integrate their database with external platforms, they won’t be able to train an ML model.
SQL Editor complexity hinders usability: Users find the SQL editor too complex to use, with effectiveness depending heavily on their SQL skills and understanding of ML concepts.
User dissatisfaction due to unmet needs: The core issue lies in user dissatisfaction, as they fail to understand the product’s value and feel that it doesn’t meet their needs.
Key Solution Elements
We focused on user needs to guide design decisions, creating a clear and consistent framework for the team. This approach encouraged innovation, improved usability, and led to a polished, trustworthy interface that strengthened the overall user experience.
Design Principles
Step by step model training: Provide users with a simplified, guided flow to help them easily build their first model.
Flexible data integration options: Allow users to upload a small test file, making it faster and easier to see results.
Start with pre-written ML Model queries: Simplify the user experience by enabling them to begin training with a pre-written ML model query.
Start with results in mind, gain value: Show users the results they can achieve before getting into the process of training a model.
Solution
To shape a solution around the identified problems, I collaborated closely with the PM and engineers to uncover any technical or business constraints. I then presented the proposed approach to stakeholders to align on direction and gather feedback.
Streamlined Workflow: A straightforward 3-step process designed to accelerate model training and deliver faster results.
Addressing the Issues: Adjusting the steps and addressing the identified issues to make sure users have a smooth and easy experience.
Value Proposition: Delivering immediate value by enabling users to achieve results more quickly.
Flowchart
My primary focus was on users as I structured the key screens of the app, aiming for a seamless and organized user experience tailored to their needs. To achieve this, I delved into researching how users reach our platform and the steps they take to get there. We have multiple landing pages with varying content guiding users through the sign-up flow and into the home page of Pecan. I explored the time taken by users to reach this point and their motivations. I sought to identify any potential frictions that could be eliminated or if there were ways to provide users with digestible information, making their initial interactions more user-friendly and not overwhelming.
Approach 1: Template gallery
Instead of directing users straight to the SQL editor, where they would need to write extensive queries from scratch and manually connect all attributes, we provided a curated list of pre-written templates tailored to the most common business use cases.
Pros Easier for users to get started by eliminating the need to write large chunks of code, preventing them from feeling overwhelmed.
Cons Users may find it challenging to select a template that perfectly matches their specific needs. They will still need to make adjustments and map their data in the code.
Approach 2: Generative chat
Similar to a template gallery, this approach also offers users pre-written templates. However, unlike a typical gallery, it allows users to type and generate a template tailored precisely to their specific needs.
Pros Compared to the first approach, users aren’t limited to a predefined list and can create templates tailored to their exact needs. It can also hold a great research value.
Cons Some users may not feel prepared or confident enough to rely on a chat tool for selecting a template they need. Also it can be price heavy and take time to develop.
Phases
Phase 1: Utilize the template gallery and evaluate the impact of simplification and the solution through key performance indicators (KPIs)
Phase 2: Leverage the template gallery and a simple version of the chat, where responses are linked directly to the template list.
Phase 3: Fully transition to generative chat as a tool for creating templates tailored to specific user needs..
Step-by-step flow
Direct all new users to a quick-start guided flow, avoiding intense laboring in SQL editor and enabling them to easily begin with Pecan and see results within minutes.
Quick Users can train their ML models within minutes and access their predictive training set output table swiftly.
Encouraging
Users who know their data will extract more value from the predictive model output.
Design Handoff
I collaborated closely with both web and mobile engineers to validate the design and copy implementation. In addition to our daily meetings, we conducted weekly demos and occasionally sit side by side to cross-check the implementation against the design intent.
Next step
Observing a notable rise in platform engagement and a substantial increase in meaningful events within new user sessions, we determined it was time for the next phase. While developing the Generative chat approach, I envisioned the entire quick start flow unfolding within the chat. This guided experience could be finely tuned to cater to the unique business needs of each user.
Comments